Loops House

Why we're building Loops House

Something odd has happened to hackathons in the last two years.

They still look the same from the outside. Sponsor booths, opening ceremonies, red-eye coding sessions, a demo day at the end. The photos on LinkedIn haven't changed. Builders show up, teams form, projects get pitched, winners get announced.

But if you look at what's actually happening inside those 48 hours, almost nothing works the way it used to.

What broke

The old hackathon assumed a specific kind of builder, someone who reads documentation before writing code, who spends the first four hours orienting themselves to a sponsor's SDK, who pitches their project on demo day because a human judge needs to be convinced. That builder still exists, but they're now a minority.

The typical builder in 2026 opens AI powered IDE, describes what they want, and starts iterating. The AI writes most of the code. Documentation gets consulted through a chat interface, not read linearly. Whether a project is any good is decided long before the pitch, it's decided by how well the builder engineered around their AI agent and whether the integration actually works.

The old hackathon infrastructure doesn't know how to see any of this.

Sponsors get told "eight projects used your API", they can't tell if those integrations were hello-world calls or production-shaped systems. They can't see which documentation gaps blocked builders, or which of their features nobody knew existed. They pay real money and get vanity metrics back.

Judges review demo pitches. But a fluent pitch and a solid codebase are now completely uncorrelated. Some of the best builders we have watched can barely present. Some of the worst ones give spectacular demos of code that wouldn't survive a single production request.

And builders themselves get nothing durable. A hackathon ends, the project page goes stale, and the work vanishes. There's no persistent surface where their actual engineering craft is visible to future employers, future collaborators, or future selves.

Every part of the system was designed for a world that no longer exists.

What we're doing about it

Loops House is a hackathon and buildathon platform designed from scratch for how builders actually work now.

The core mechanic is simple: sponsor documentation becomes an AI agent that lives inside the builder's IDE. Instead of a wiki nobody reads, sponsor context is a Skill that Cursor, Claude Code, and Codex can invoke directly. When a builder needs to understand how a webhook works, the answer comes to them, in their editor, in the flow of building.

Every project gets AI-judged based on evidence, the actual code, the actual test results, the actual integration quality. Not the polish of the demo. Evaluations are interrogable: a sponsor can ask "did they actually integrate our SDK, or just call it once? Show me the code" and get a real answer. What gets built and how it got there both stand.

Every project also gets a permanent, SEO-optimized page on Loops House. A living page that builders can point to for years, that shows up when their name gets Googled, that credits their work as their own portfolio.

For sponsors, we produce telemetry that no other platform can: real interaction data on how builders engaged with their SDK, where builders got stuck, which of their documentation gaps blocked the most people, and which builders shipped the highest-quality integrations with the fewest iterations. Sponsors don't just pay for logo placement anymore, they get insight they'd otherwise pay a DevRel team six figures to collect.

We've run this at real events. NextGen BioAgents at NY Tech Week 2026, hosted by Nucleate, sponsored by Pfizer, Nebius, Cursor, and Fulcrum Science where builders shipped 23 projects in one day, using AI Skills for each sponsor's stack. Codex Community Hackathon Pune saw over 1,500+ registrations, 300+ selected builders with 85 projects scored live on a public AI evaluation board.

What we saw at both events convinced us this isn't just a better hackathon. It's a different category.

Why it matters

A few things become possible when hackathons stop being pitch competitions and become craft demonstrations.

Builders with real engineering ability but weak presentation skills stop losing to teams with the opposite profile. Domain experts, the biotech researcher who codes on the side, the clinician building tools for their own practice, start winning, because the judging looks at the work, not the theater.

Sponsors stop treating hackathons as marketing spend and start treating them as product development inputs. When you can see exactly where builders got stuck in your SDK, hackathon sponsorship becomes cheaper than user research and more honest than support tickets.

Hiring companies get a new signal. Traditional resumes are collapsing under the weight of AI-assisted work, nobody knows how much of "I built X" was really the candidate. But a Loops profile shows how the builder worked: how many iterations they needed, how they structured their AGENTS.md, how efficiently they used their AI agent. That signal is much harder to fake than a GitHub commit history.

Domain-specific hackathons become viable at scale. Biotech-AI, legal-AI, climate-AI — verticals that couldn't previously justify their own hackathon circuits — can now run events where domain expertise gets rewarded and sponsors get precise feedback on how their tools land in specific industries.

None of this requires convincing anyone that AI is going to change software. That argument is over. What's still contested is what the infrastructure for AI-native builders should look like. Loops House is our answer.

Where we're going

Three arms, one platform.

Hackathons

This is where we started. Where sponsor SDKs become agents in builders' IDEs, projects get AI-judged on evidence, and every project ends up with a permanent home.

Compete

Think Build Battles. Three or Five projects, ten hours, teams and solo. Live leaderboards. Rating tiers from Builder to Grandmaster. What Competative Coding platforms are for algorithmic puzzles, Compete is for shipping software.

Grind

A growing library of problem statements organized into skill categories: AI Voice, RAG and Retrieval, Agentic Engineering, Vertical AI in biotech and legal and beyond. Solve enough problems in a category and earn credentials that live on your profile.

The vision underneath all three is the same, make the discipline of building software with AI visible, portable, and rewarded. Every builder should have a public, evidence-based record of what they can actually do. Every sponsor should be able to see how their tools land in real hands. Every domain should have a way to grow its next generation of AI-native builders without waiting for permission from Silicon Valley.

We're building the infrastructure for the era that's actually here.

---

What's next

We're running events. We're growing our sponsor list. We're onboarding communities and universities. If you host hackathons, sponsor them, or just build in them, we'd love to talk.

Loops House is at loops.house.

Come build with us. The Loops House cat will be waiting. 🐈